Documentation Required

[uzair004]

Please put some effort in writing version specific documentation for NodeJS weaviate-client pacakge.

I have tried every single AI tool I can think of

• ChatGPT
• Gemini
• Perplexity
• DeepSeek
• Grok
• Copilot
• Claude

and none of them can get the syntax right, they can't even find the details of the new 3.x version for some reason.

Also, the documentation is bit vague and hard to read.

The simple stuff like to How to use gemini models for embedding are not well documented, it has 4 different headers and none of them work, had to shift to default weaviate embeddings.

[bevzzz]

cc @malgamves

[malgamves]

Hey @uzair004, sorry you were having trouble with the documentation.

Is there anything specific you would like us to document that you didn't find? What parts did you find vague and hard to read?
On LLMs referencing our client, I tried asking Claude Sonnet 4 to make a RAG pipeline with the v3 client and got this back. What model did you have trouble with?

Sadly we have no say on what data the models train on but do our best to make sure the docs are up to date.

The simple stuff like to How to use gemini models for embedding are not well documented, it has 4 different headers and none of them work, had to shift to default weaviate embeddings.

I know how frustrating that can be, sadly as Google released models under Vertex, then AI Studio, etc, we've had to support and deprecated multiple headers and add additional ones. I understand we missed a few. Please tell me which ones didn't work for you. I'll fix them as soon as I can.

[uzair004]

Well, none of the services I mentioned can get it correct.

As of today, the chatgpt, claude and gemini and copilot, cursor struggle to find out how to do upsert behaviour on weaviate operations.

Each of those suggest using upsert method that doesn't exist in weaviate-client@3.8.
Also suggest that replace method would do the trick and insert should update the record if already exist.

Surprisingly despite having same namespace and id for uuid generation using generateUuid5 provided by weaviate-client. It always leads to duplicate records in weaviate.

For previous connection related code, the LLMs and assistance tools still generate the local connection style code instead of using connectToCloud method per the package.

[uzair004]

Update: The LLMs and AI Assistance tools (Cursor, CoPilot) could barely generate any useful code for weaviate npm 3.8 version.

The Weaviate inbuilt AI Chat is decent but it is not robust enough to provide accurate code and find bugs in weaviate related code.

LLMs either generate non-existing methods for weaviate, get their definition and usage wrong or couldn't differentiate what to pass.

[uzair004]

Hey @uzair004, sorry you were having trouble with the documentation.

Is there anything specific you would like us to document that you didn't find? What parts did you find vague and hard to read? On LLMs referencing our client, I tried asking Claude Sonnet 4 to make a RAG pipeline with the v3 client and got this back. What model did you have trouble with?

Sadly we have no say on what data the models train on but do our best to make sure the docs are up to date.

The simple stuff like to How to use gemini models for embedding are not well documented, it has 4 different headers and none of them work, had to shift to default weaviate embeddings.

I know how frustrating that can be, sadly as Google released models under Vertex, then AI Studio, etc, we've had to support and deprecated multiple headers and add additional ones. I understand we missed a few. Please tell me which ones didn't work for you. I'll fix them as soon as I can.

I have tried most of the headers for the Gemini API, my API key was generated in Vertex. Had to switch to weaviate default embedding model unfortunately.